PolynomialCountSketch.bitHash_ docstring says dtype=float32 but is actually int64 #34923

Introduce yourself

I use scikit-learn for general ML work. I found this while cross-checking
the PolynomialCountSketch docs against its implementation for an unrelated
performance PR (#34919) and noticed the bitHash_ attribute's documented
dtype doesn't match what the code actually produces.

Describe the bug and give evidence about its user-facing impact

PolynomialCountSketch's bitHash_ attribute is documented as:

bitHash_ : ndarray of shape (degree, n_features), dtype=float32
    Array with random entries in {+1, -1}, used to represent
    the 2-wise independent hash functions for Count Sketch computation.

but it is actually set with:

self.bitHash_ = random_state.choice(a=[-1, 1], size=(self.degree, n_features))

numpy.random.RandomState.choice with an integer array a=[-1, 1] and no
explicit dtype produces an integer array, not float32. This has been the
case since the class was introduced (#13003, 2019),
so it is not a regression, just a docstring that was never accurate.

User-facing impact is limited (the attribute is internal-ish and the class
still works correctly), but anyone relying on the documented dtype — for
example writing code that assumes bitHash_.astype(np.float32, copy=False)
is a no-op, or reasoning about memory usage from the docstring — would be
misled.

Steps/Code to Reproduce

import numpy as np
from sklearn.kernel_approximation import PolynomialCountSketch

X = np.random.RandomState(0).random_sample((50, 20))
ps = PolynomialCountSketch(degree=2, n_components=10, random_state=0).fit(X)
print(ps.bitHash_.dtype)

Expected Results

float32

(matching the docstring)

Actual Results

int64

Versions

System:
    python: 3.14.6 (v3.14.6:c63aec69bd5, Jun 10 2026, 08:07:54) [Clang 21.0.0 (clang-2100.1.1.101)]
executable: /Library/Frameworks/Python.framework/Versions/3.14/bin/python3
   machine: macOS-26.5.2-arm64-arm-64bit-Mach-O

Python dependencies:
      sklearn: 1.10.dev0
          pip: 26.1.2
   setuptools: 83.0.0
        numpy: 2.5.1
        scipy: 1.18.1
       Cython: 3.3.0
       pandas: None
   matplotlib: None
       joblib: 1.6.0
threadpoolctl: 3.6.0
     narwhals: 2.25.0

Built with OpenMP: True

I'd be happy to open a PR to fix the docstring to say the actual dtype
(int64, or more precisely whatever numpy.random.RandomState.choice
returns for an integer input array) once this is triaged.